Published on the 29/01/2025 | Written by Heather Wright
Tech news from the WEF’s annual summit…
AI dominated talk at Davos 2025 as the World Economic Forum’s annual meeting, with its theme of collaboration in the intelligent age, took over the Swiss ski resort, and none more so than the chatter about China’s DeepSeek.
The ‘breakthrough’ open source DeepSeek 3.0 AI language model, created by the Hangzhou-based startup of the same name, quickly caused market turmoil, sending shockwaves through investors and causing consternation among some who questioned whether China – despite export controls imposed by the United States, is now surpassing the US in AI.
“That is the power of open research and open source.”
The DeepSeek model shot to the top of the app store downloads, and caused the stock prices of some tech vendors, such as Nvidia to sink (briefly) as investors rapidly adjusted their bets in the face of the low-cost alternative – DeepSeek has claimed training costs of ‘just’ $6 million for the model, rather than the billions spent by others. Nvidia’s stocks slumped, with the company initially shedding US$600 billion in value – the biggest one-day loss for a. company in stock market history.
Despite the lower training costs, the R1 model is reportedly comparable in ability to the best models from companies including OpenAI, Anthropic and Meta. It’s also built on less powerful chips thanks to US export restrictions which prevent the export to China of cutting edge tech. DeepSeek says the most recent model was trained on the less powerful Nvidia H800s GPUs specifically developed for the Chinese market and with speeds capped at half that of Nvidia’s top offerings.
The company, founded in July 2023 by Liang Wenfeng, has reworked its training process to reduce strain on the GPUs, and uses a ‘chain of thought’ approach, similar to that of ChatGPT, to break down complex problems into small, logical steps, and processing queries step by step.
Much was made of what DeepSeek means for China’s future in AI – and that of the US.
China’s Next Generation AI Development Plan, launched in 2017, includes a bold goal to become a global leader in AI by 2030. It also positions AI as a core driver of economic transformation by 2025.
But Meta VP and chief AI scientist Yann LeCun says the question isn’t whether China is surpassing the US in AI, instead ‘the correct reading is: Open source models are surpassing proprietary ones’.
In a LinkedIn post he says DeepSeek has profited from open research and open source, including Llama from Meta.
“They came up with new ideas and built them on top of other people’s work. Because their work is published and open source, anyone can profit from it. That is the power of open research and open source.”
Tech stocks have steadied since their earlier DeepSeek-inspired slump, but the technology has raised questions about whether demand for high-end chips such as those from Nvidia and Dutch company ASML, which also took an early hit, could be affected, along with questions over whether investors are overvaluing stocks buoyed by the promise of AI.
But the offering is also raising questions about some of its erratic responses – and the issue of what information has been censored – don’t expect to find out about the Tiananmen Square uprising, or the Chinese government’s harsh response here, with the app reportedly censoring itself in real time – along with privacy concerns.
Earlier this week Australia’s Minister for Industry and Science Ed Husic questioned whether Chinese products had same appreciation of data and privacy management that export markets would require.
DeepSeek’s own privacy policy notes it collects information automatically such as technical, user and payment information and cookies and information from other sources such as Apple or Google log-ins used to access the service, and data from advertising, measurement and ‘other’ partners, along with data provided by users themselves, such as profile information and user inputs and chat histories. Data, it says is stored in ‘secure servers’ in China.
Beyond DeepSeek
DeepSeek wasn’t the only AI talk at Davos.
Agentic AI and AI agents, including OpenAI’s Operator, were a hot topic, with their ability to automate tasks, and chain together tasks to create workflows.
SAP and Salesforce were both talking up their agent plans. SAP will launch two agents this year, one for sales and one for the supply chain. The sales offering will recommend the best price and how to bundle products to hit the consumer at the right point in time, while also working with the supply chain agent to ensure product is in stock and will be delivered on time.
One caveat however: SAP admits about 80 percent of its customers don’t have the infrastructure, team or investment to implement AI agents. SAP says it will make it easy for customers through including the predictions and automation in the software.
Smart factories – whether via AI powered autonomous production lines, digital twins simulating entire supply chains, or next-gen robots were also among the discussion points. One session saw industry leaders discussing the impact of such technologies in enabling organisations to address issues including supply chain resilience and climate change.
Crypto also made its presence felt, with discussion around how policy changes could impact the financial system, while the impact of cyber incidents was also noted in a session discussing responsible approaches to building and maintaining trust before, during and after a cyber crisis.
